.png){kind=icon}
.png){kind=icon}
Return
Robust feature selection with trend fuzzy neighborhood rough sets for multivariate long-term forecasting
.png){kind=icon}
PRE
AI
C
X
Z
DOI：10.1016/j.fss.2026.110037.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Fuzzy neighborhood rough sets (FNRSs) are effective for feature selection under uncertainty. However, applying them directly to multivariate time series forecasting disrupts temporal order and suffers from noise sensitivity. To address these issues, this paper proposes Trend Fuzzy Neighborhood Rough Sets (TFNRSs), a feature selection method specifically designed for long-term multivariate time series forecasting. Unlike conventional FNRS that rely on fixed radii and raw distance-based similarity, TFNRS adaptively circumscribes neighborhoods using linear fuzzy information granulation and measures similarity by comparing underlying data trends rather than raw distances, preserving temporal structure and enhancing robustness. To further extract accurate trends, we propose Neural Adaptive Trend Filtering (NATF), an end-to-end learnable method that dynamically optimizes the granulation process. Experimental results on real-world datasets demonstrate that TFNRS selects significantly more compact feature subsets and achieves superior long-term forecasting accuracy compared to state-of-the-art methods.
Journal
IF:
2.7
Papers:
7.6K
Citations:
1.5W
Organization
Cited Papers